> ## Documentation Index
> Fetch the complete documentation index at: https://docs.koalr.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Limits and security

> Rate limits, the read-only guarantee, permission scoping, and how untrusted content is handled.

## Rate limits

| Limit | Value                                           |
| ----- | ----------------------------------------------- |
| Burst | 60 tool calls per minute, per user              |
| Daily | 1,000 tool calls per rolling 24 hours, per user |

Over the limit, the tool returns a message saying so and when it resets, with HTTP 429 at the transport level. Limits are per user, not per organisation, so one agent cannot exhaust an agency's whole team.

## Read-only, always

Every tool reads. Nothing an agent does can change a site, edit prompts, start a run, create a card, generate a draft or spend anything on your account. Kevin's writing capabilities in the app are deliberately withheld from MCP.

## Your permissions, nothing more

Every call runs as the signed-in user. Workspaces you do not belong to and sites you are not assigned to do not exist as far as the tools are concerned. Membership is checked live on every call.

## Untrusted content

Answers, competitor pages and citation snippets are written by third parties. Koalr labels them as data to report on, so an instruction hidden in a competitor's page is not followed by your agent. Tools that return excerpts (`get_prompt_detail`, `get_citations`) carry the warning in their descriptions so well-behaved agents apply it automatically.

## Disconnecting

Remove the server from your client and the agent stops immediately. Sign-ins expire within 24 hours; if you need a connection cut off sooner, email support and say which client you used. Membership is checked live on every call, so removing someone from a workspace stops their agent within seconds.

## Reporting a problem

Email [support@koalr.ai](mailto:support@koalr.ai) with the client, the tool and the time. Include the request identifier from the error if there is one.